The Startup Ideas Podcast
The Startup Ideas Podcast

Jack Dorsey's Buzz: The New Hermes Agent?

July 28, 2026

AI Summary

5 min read

Jack Dorsey’s new tool, Buzz, is being called a “Slack killer,” but the real story is more interesting than that. It’s a team chat app where AI agents are first-class members, not add-ons. Built on an open protocol called Noster, Buzz lets you switch between AI models and harnesses without losing context, share compute with your team, and build and deploy software directly from a conversation. For solopreneurs and small teams, it offers a glimpse into a future where your agents are as much a part of the team as your human colleagues.

The Core Idea: Agents as Teammates, Not Integrations

The fundamental difference between Buzz and a tool like Slack is that agents are native to the experience. In Slack, you might add a bot or an integration. In Buzz, agents are members of your team. They have names, they sit in channels, and they participate in conversations. This shift changes how you think about work. Instead of asking a tool to do a task, you ask a teammate.

Continue reading the full summary in the app — free to try.

Read Full Summary →

Free • No credit card required

What you'll learn

  • 1 (00:00) **Buzz Launch and Episode Preview** - Jack Dorsey's new app Buzz is introduced as an "agentic version of Slack," and the episode promises to explain what it is, why it matters, and how to use it.
  • 2 (01:06) **Guest Introduction and Episode Goals** - Guest Vinny joins to break down Buzz, committing to share best practices for using agents to get things done.
  • 3 (02:23) **The Core Selling Point: Openness** - Vinny argues that Buzz's real advantage is its open architecture, not just being "Slack with agents."
  • 4 (03:31) **Swappable Agent Harnesses: A Game Changer** - The most powerful feature is the ability to swap the underlying AI model or "harness" under any agent without losing context.
  • 5 (05:25) **Why This Matters for Productivity** - The swappable harness feature acknowledges how people actually work in the fast-moving AI landscape.
  • 6 (06:25) **Global Skills and Agent Portability** - Agents can access globally installed skills and files, which persist even when changing harnesses or models.
  • 7 (06:55) **Audio Huddles with Agents** - Buzz allows live voice conversations with agents through audio huddles, enabling more natural, creative collaboration.

+ Full timestamped outline available in the app

Show Notes

I sit down with Vinny for a live tour of Buzz, an open source, agent-native chat app from Block built on an open protocol. Vinny makes the case that openness is the real story here: agents arrive as first-class teammates, the harness underneath each agent swaps freely between Claude Code, Codex, Goose, and open code, and your entire chat context travels with you through every swap. He demos real output, including a CRM app built with the Wasp full stack framework and deployed to Railway, plus a tweet leaderboard that pipes daily stats back into a channel through a public API. I press him for the honest state of the software and for the setup advice he actually uses day to day. By the end I share where I land on Buzz versus Slack, and why I think anyone building right now gains from putting their hands on tools like this.

Timestamps

00:00 – Intro

02:57 – Agents as First-Class Team Members

03:49 – Swappable Harnesses Under Any Agent

06:55 – Audio Huddles With Agents

08:34 – Git, Feature Branches, and Parallel Worktrees

11:20 – Building a CRM App with Buzz and Agents

13:43 – Why This Matters

18:13 – Best way to engage with your Agents

23:53 – Shared Compute and Local Models

25:42 – Model Choice, Data Ownership, and Lock-In

27:36 – Context as the Foundation

29:39 – Setting Up Agents

31:21 – Skills and Speed

33:03 – Who Should Try Buzz Today

34:36 – My Take: Live in the Future

38:10– Closing Thoughts

Key Points

  • Buzz treats agents as members of your team, so one shared chat becomes the context layer for humans and agents together.
  • The harness under any agent swaps freely, and every chat, project, and decision comes along for the ride.
  • Globally installed agent skills stay available inside Buzz, so an existing Claude Code setup carries straight over.
  • Agents branch, work in parallel worktrees, push to Git hosting on your own relay, and ship live apps end to end.
  • Shared compute lets a small team run one local model on one machine and use it from many computers.
  • Buzz sits in early preview today, which makes it a strong fit for solopreneurs and small teams iterating fast.

The #1 tool to find startup ideas/trends - https://www.ideabrowser.co

The Startup Ideas Podcast